Awesome RGB-D Datasets

This repository contains the selected list of datasets found in our survey "A Survey on RGB-D Datasets". We gathered 231 datasets that contain accessible depth data, therefore, this is the criteria to be considered an awesome dataset!

Datasets are divided into three categories and 6 sub-categories, which represent distinct applications of RGB-D data. The taxonomy tree of the application types is available in Figure 1, and extra information and examples of each category are available in our paper.

Contents

1. Information
2. Update
3. Citation
4. Website
5. Contributing
6. Datasets

Information

We present each dataset with 8 columns that summarize important information of the datasets: Year, Scene Type, Sensor Type, Sensor Name, Data Modalities, Extra Data, Images/Scenes, and Application.

In Figure 2, we illustrate the variability of datasets presented in this paper, and how the type of sensor produces distinct depth results, changing reliability and sparsity.

Data is organized here by "Application Type", "Scene Type", and "Year" in this order. Since it is a long list, we also have it available on our website in a filterable way.

We also discuss the different applications for each sensor type and explain how these sensors work in our paper. We also identify influents and trending datasets in each field, which are also detailed in our paper.

Update

15/07/2022 : Added 28 new datasets after revising other 606 papers. The majority of datasets included in this revision contain saliency maps and are from the "Body" category.

Citation

Please cite our paper if you used our survey in your work:

@article{LOPES2022103489,
title = {A survey on RGB-D datasets},
journal = {Computer Vision and Image Understanding},
volume = {222},
pages = {103489},
year = {2022},
issn = {1077-3142},
doi = {https://doi.org/10.1016/j.cviu.2022.103489},
}

Website

We made available a website with filtering and ordering options to help you find the best datasets for your work. We encourage you to check it out in our page.

Contributing

If you have any comments or want to add your work to the list, just contact us at:

• alexandre.lopes at ic.unicamp.br or
• alexrlops at gmail.com

We expect to continue updating this list of datasets through the years, and your contribution is fundamental to continue this work.

RGB-D Datasets

No.	Dataset Name	Sensor Type	Sensor Name	Application Type	Scene Type	Data Modalities	Extra Data	Images/Scenes	Year
0	No Name Defined	-	Synthetic	SOR, and SOE	Aerial	Color, Depth	Normal Maps, Edges, Semantic Labels	15 scenes (144000 images)	2020
1	DDAD	LiDAR	Luminar-H2 LIDAR	SOR, and SOE	Driving	Color, Deph	Instance Segmentation	150 scenes (12650 frames)	2020/2021
2	Woodscape	LiDAR	Velodyne HDL-64E	SOR, and SOE	Driving	Color, IMU, GPS, Depth	instance segmentation, 2d object detection	50 sequences (100k frames)	2019/2020/2021
3	NuScenes	LiDAR	-	SOR, and SOE	Driving	Color, Depth, Radar, IMU	3D object detection, semantic segmentation	1000 scenes (20 seconds each). 1.4M Images and 390k Lidar Sweeps	2019/2020
4	EventScape	-	Synthetic	SOR, and SOE	Driving	Color, Depth	Semantic Segmentation, Navigation Data (Position, orientation, angular velocity, etc)	758 sequences	2021
5	KITTI-360	SCS, LiDAR	Velodyne (LIDAR) Points Cloud, MVS	SOR, and SOE	Driving	Color, Depth, GPS, IMU	2d-object detection, 3d-object detection, tracking, instance segmentation, optical flow. These are not in necessary in the same dataset	11 sequences to over 320k images and 100k laser scans